Atlas

Models

← All models

GPT-5.5 Pro

OpenAI · 2026-04-23

GPT-5.5 Pro is OpenAI's higher-compute GPT-5.5 variant, publicly released in ChatGPT on April 23, 2026 alongside GPT-5.5 Thinking; OpenAI describes it as the same underlying model using a parallel test-time-compute setting. It is designed for harder questions and higher-accuracy work, is available in ChatGPT to Pro, Business, and Enterprise users, and became available through the API on April 24, 2026 as gpt-5.5-pro, with medium, high, and xhigh reasoning-effort settings.

Benchmark scores

BenchmarkScore
ARC-AGI-1 Semi-Private96.5
ARC-AGI-284.2
BrowseComp90.1
BullshitBench v2 Leaderboard36.0
BuseyBench SVG6.1
Capability163.1
Chess Puzzles64.0
Chess Puzzles (task version 2.0)62.0
CritPt (avg@5)30.6
DTBench96.0
Epoch Capabilities Index (ECI)163
FrontierMath Tier 4 (v1) - Private39.6
FrontierMath Tier 4 (v2) - Private78.0
FrontierMath Tiers 1-3 (v1) - Private52.4
FrontierMath Tiers 1-3 (v2) - Private87.7
GDPval82.3
GeneBench - release-reported version unspecified33.2
GeneBench-Pro20.5
GeneBench-Pro Artificial Analysis Subset9.2
GeneBench-Pro Externally Reviewed18.0
GeneBench-Pro Not Externally Reviewed24.7
GeneBench-Pro Public Release Subset14.0
GPQA Diamond93.9
Humanity's Last Exam43.1
Humanity's Last Exam (with tools)57.2
LMCA53.9
MineBench — Bradley-Terry Rating1938
MineBench Conservative Rating1883
OpenAI investment-banking modeling - internal release score88.6
OTIS Mock AIME 2024-2025100.0
Loading Atlas data…